Research On Session Recommendation Method Based On Graph Neural Networks

In recent years,with the continuous development of artificial intelligence and Internet technology,people’s lives are gradually filled with a variety of information,and recommendation systems can help people filter the right information from the huge amount of information and recommend it to users.Traditional recommendation methods recommend items to users based on their needs by modeling their behavioral data.However,anonymous users’ personal information and long-term historical behavioral data are not available to traditional recommendation methods.The session-based recommendation method can analyze the interests and preferences of anonymous users based on their behavior over a recent period of time and recommend items to them more accurately.In this paper,the problems of insufficient utilization of session information and insufficient access to item-item reliance relations in the session-based recommendation method are thoroughly investigated,and the main work is as follows:(1)To address the problem that the directed session graph does not fully utilize data of the order and numbers of clicks on items in a session,a relationship-enhanced session graph incorporating session context information and a relationship-enhanced matrix storing information on the structure of the corresponding graph are constructed in this paper,and a session recommendation method SC-SRM that utilizes session context information is proposed in this paper.The improved relationship-enhanced session graph can use the graph structure to obtain richer information relationships between items,while differentiating the importance of different items in a session.The method captures the dynamic preferences of users through a combination of long-and short-term interests and uses them to make subsequent recommendations.(2)To address the problem of insufficient utilization of dependencies in a session,the combination of the multi-layer self-attention mechanism and the soft-attention mechanism is used to capture interest dependencies between items throughout the session sequence.In this paper,based on the proposed SC-SRM method,the attention mechanism-based session recommendation method AM-SRM is further proposed.The multi-layer self-attention mechanism is capable of adaptively attending to the relevance of all elements to each other and mitigate the problem of gradient disappearance caused by multi-layer neural networks by adding residual connection to the method.The combined use of the two attention mechanisms can improve the recommendation effectiveness of the method by taking into account the interest relationships that exist between other items,while ensuring the association between all items in the session and the last item.Finally,in this paper,two methods are proposed for comparing and ablating with the comparison methods on two publicly available datasets,Diginetica and Yoochoose1/64,and P@20 and MRR@20 are selected as the evaluation metrics for the experiments in this paper.The experimental results show that the two proposed methods in this paper have improved on the two evaluation metrics,which fully verifies the effectiveness of the methods.